[quintesse]

• parse separator continuations with pending backslash states
• decode Unicode escapes after removing line continuations
• reject malformed and non-ASCII Unicode escapes
• preserve blank and comment lines after empty continuations

These were all corner cases where Codejive Properties did not behave like Java Properties.

To make these fixes possible the core parser loop was refactored somewhat to enable a more extensive state machine.

[quintesse]

@wleeaf I would be interested to know how these fixes handle your extensive test suite. You think you'd be able to test it before I create a new release?

[wleeaf]

I ran the differential corpus against main (7a247ef): 20,061 inputs (63 hand-written edge cases plus random mixes of escapes, \uXXXX, line continuations, comments, CR/LF/CRLF line endings and Latin-1 bytes), each loaded with java.util.Properties (JDK 21) and with codejive Properties, plus a store() and JDK reload of the codejive result.

• main: no differences. All 14,760 inputs that the JDK loads give the same keys and values, and store() reloads to the same map. The 5,301 inputs with a malformed \u escape now throw IllegalArgumentException, like the JDK.
• No regressions: no input that matched on v0.0.8 or on main right before A set of regression fixes where we differed from Java's Properties #100 differs now.
• Right before A set of regression fixes where we differed from Java's Properties #100, 1,091 inputs differed (for example \u00a\ + newline + a=b threw instead of giving the key ª, and a lone \ line with CRLF produced an empty key), and the malformed escapes threw RuntimeException. On v0.0.8 it was 3,553 inputs, plus the same exception type difference.
• A second corpus of the same size with a different seed gives the same result: no differences on main.

[quintesse]

So you are okay with creating a new release, right?

Or do you see any other issues that still need to be fixed?

[wleeaf]

Nothing outstanding from my side. The differential shows no differences from java.util.Properties on main, and the only cases I had noted, the lone \ line and the continuation inside a key, are covered by #98 and #100. Fine to release as far as I can see.

[quintesse]

OK, version 0.1.0 released!